Argilla

Argilla is an open-source collaboration tool from argilla-io for AI engineers and domain experts who need to build high-quality datasets. It combines a web annotation UI with a Python SDK and server, and can be deployed on Hugging Face Spaces or with Docker. The latest listed release is v2.8.0.

What It Is

Argilla is a data annotation and human-feedback platform for AI projects. It is used to collect feedback for traditional NLP tasks such as text classification and named entity recognition, for LLM work such as RAG and preference tuning, and for multimodal models such as text-to-image. Teams define datasets with fields, questions, guidelines, metadata and vectors, then log records and have annotators respond through the UI.

How the Workflow Works

Install the Python SDK with pip and deploy the Argilla server. Connect with the Argilla client using an API URL and key, define Settings (for example a text field plus a label question), create the dataset, and log records, for instance from a Hugging Face dataset. Annotators then label records in the UI, with filters, AI feedback suggestions and semantic search to speed up the work. The docs also cover user and workspace management, task distribution, querying and filtering, import and export, custom fields, Markdown content and webhooks for server events.

Examples from the Docs

The documentation describes community datasets and models built with Argilla, including a cleaned UltraFeedback dataset and a distilabel-improved Intel Orca DPO dataset. It also references presentations from the Red Cross, Loris.ai and Prolific describing their own use of the tool.

Current Status

The README states that the original authors have moved on to other projects. The codebase is described as mature and stable, with no new features planned. Bug fixes and patches will continue to be published as needed, and the maintainers are looking for new maintainers.